| name | varlens-verify-before-done |
| description | Use when about to claim a VarLens change is done, working, fixed, or ready to commit/PR — to pick and run the right make target for what you changed and report its actual output. Symptoms of about to skip this: "the diff looks clean", "types should pass", "this is a trivial change", "I'll let CI catch it", or claiming a UI change works without opening the app. |
VarLens: verify before claiming done
The rule
Run the relevant make target and report its actual outcome. Never infer success
from a clean diff or from "it should work". Evidence before assertions. If you can't
run verification in this environment, say so explicitly — don't imply it passed.
This is the repo's own top gate (make ci is the local minimum; make ci-full is the
canonical go/no-go). For the general discipline see
superpowers:verification-before-completion; this skill adds the VarLens command matrix.
Which target for which change
| Your change touches | Run (minimum) |
|---|
| Any code at all | make typecheck |
| Anything you'll commit | make ci (lint-check + format-check + typecheck + rebuild-node + test) |
| Renderer, IPC, database, or workers | make rebuild-node && make test (then make ci) |
| Electron lifecycle, packaging, worker bootstrap, native module | make ci-full (includes startup smoke + packaged smoke) |
| A UI component | Build/run the app and look at it — make dev or /run. A diff that typechecks is not a working UI. |
make ci deliberately serializes the heavy gates to bound peak memory. On this machine,
wrap heavy runs in a memory cap: systemd-run --user --scope -p MemoryMax=16G make ci-full.
Order of operations (the native-ABI trap)
Tests run on the Node ABI; the app runs on the Electron ABI. make ci /
make test need make rebuild-node first, or the SQLite module ABI-fails and the run is
invalid — not a pass, not a real fail. If tests won't load out/main/db-worker.js or you
see NODE_MODULE_VERSION errors, that's ABI, not your code — see varlens-native-rebuild.
Reporting
State what you ran and what it printed:
- ✅ "Ran
make ci — lint, format, typecheck, and the full test suite passed (report the count make test prints)."
- ✅ "
make test fails: useFilters.test.ts expects the new field. Fixing the test."
- ✅ "Typecheck passes. Could not run
make ci-full here (no Electron display); startup smoke unverified."
- ❌ "Done — the change looks correct." (no command run)
- ❌ "This should pass CI." (prediction, not evidence)
Never
- Never lower a coverage / lint / typecheck threshold to make a red suite green. Add
tests or fix the code. Lowering the bar is a cover-up, not a fix. (Baselines like
scripts/agent-health-baseline.json are not to be relaxed to fit new code either.)
- Never claim a UI change works without opening the app. Screenshot or drive it.
- Never report "tests pass" when they were skipped, or when the ABI was wrong. A
skipped or ABI-failed run is not a pass — say which happened.
